xe: sdpa: fix several errors related to out of bound accesses in sdpa #2497
+73
−25
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
umar456
added
bug
umar456
force-pushed
the
uarshad/fix_sdpa_config_2nd_token
branch
from
3952a04 to
cd354b5
Compare
umar456
force-pushed
the
uarshad/fix_sdpa_config_2nd_token
branch
from
cd354b5 to
9a94ead
Compare
dzarukin
approved these changes
Jan 24, 2025
|
make test |
ElaineBao
approved these changes
Jan 24, 2025
| --case=complex_fusion/mha/sdpa-compressed-kv-int4-gs32.json | ||
|
|
||
| # llama-2-7b-chat | ||
| --in-shapes=0:1x32x128x384*abdc+1:1x32x1x384+2:1x32x1x384+3:1x32x384x128+5:1x1x384x384+6:1x32x384x128+7:1x32x384x1+8:1x32x384x1,\ |
There was a problem hiding this comment.
nit: better to reset dt here, to align with other tests, and avoid misunderstanding.
syurkevi
approved these changes
Jan 24, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR addresses several out of bound access errors related to second token shapes(Q = 1 K/V = SEQ_LEN+1).
Additionally this PR fixes issues related to compilation errors when using common zero points masks are used without scales when quantizing.